Example

  
Figure 3: Interpretation of Example 2 from Martin's (A's) point of view.

We will show how our model works on Example 2 of section 1.2: in the interaction, Loes, who is denoted as B, is the receptionist and the keeper of supplies. Marty is denoted as A.

T1: A: ``Loes, do you have a calendar,''

T2: B: ``Yeah'' ((reaches for her desk calendar))

T3: A: ``Do you have one that hangs on the wall?''

T4: B: ``Oh, you want one.''

T5: A: ``Yeah''

In this interaction, there are two repairs directed towards two distinct instances of misunderstandings; the first one is a third position repair in T3 and the second one is a (different) self-repair in T4, due to the fact that B recognizes a further misunderstanding from the request for repair in T3.

Figures 3, 4 and 5 will be used in the following to describe the different interpretations of this interaction by the two speakers. In the figures, we only show the most important actions of the rightmost path in the recognized Agent Modeling plans. As a fact, the structures would contain other previous AM actions, so we only report the interpretation branch which contains the pending goals. We will now analyze the dialogue from the different points of view of the two speakers.

Figure 3 shows A's point of view: in CTX0, A has the goal of hanging a calendar in his office (``Satisfy(A, A, done(A, Hang(A, Calendar1), office(A)))").

In order to do that, he must satisfy the precondition of action ``Hang'', that is ``have(A, Calendar1)'' (see the DM containing ``Hang(A, Calendar1, office(A))" in the figure); so he starts another ``Satisfy" action on the goal of having a calendar. The execution of this AM action leads A to plan an adequate object level action to induce B to provide him with the calendar (the ``Find-plan'' AM action is not shown in Figure 3). A request is a good action for that, but it is not very polite; so, A refers indirectly to his intentions by checking whether B can execute the action (i.e. whether she has any calendars; see the action ``Satisfy(A, B, Knowif(A, have(B, Calendar1)))''). In other words, A hides his intention to execute the request by performing a question (``Ask-if(A, B, have(B, Calendar1))''), from which B can infer A's intention (this way of acting is a ``pre-request", an indirect politeness strategy described in [Levinson1983]).

We now analyze B's (Loes') view of the dialogue, shown in Figure 4. While B does not have any problem in interpreting A's question as a request for a calendar, she misinterprets her partner's high level intentions underlying the underspecified sentence (see CTX0): she thinks that A wants to borrow the calendar to have a look at it (the underlying domain action is ``Read"). As a consequence of her misinterpretation, she chooses a calendar of the wrong type, a desk one: Figure 4 shows her action by associating the (``wh'') restriction ``desk-calendar(Calendar2)'' to action ``Provide(B, A, Calendar2)'' in context CTX1.
Since B is collaborative, she adopts A's still unexpressed intention that she gives him a calendar: this fact has been denoted in the figure by means of the dotted arc that relates the effect of the request (``SH(A, B, Cint(A, B, Goal(A, Goal(B, Provide(B, A, Calendar2)))))") with action ``Satisfy(B, A, done(B, Provide(B, A, Calendar2)))''.

Going back to Figure 3, we can examine how A interprets the observation that B is taking a calendar (``Exec(B, Take(B, Calendar2))"). The interpretation heuristics mentioned in section 3.2 are used to perform the interpretation task: instead of identifying all the possible reasons for taking a calendar, the interpretation process tries to find a relation between A's goals and the new contribution. Starting from the low-level goals, taking something has nothing to do with A's explicitly expressed goal of being informed about something; however, the heuristics relate A's implicit request that B provides him with a calendar with the fact that B has just taken one: to provide an agent with something, you have to take the thing and pass it to the receiver.

  
Figure 4: Interpretation of Example 2 from Loes' (B's) point of view.

Unfortunately, B's action can not be related with A's dialogue context CTX0 because B is handling the wrong type of calendar; so, the coherence-seeking procedure fails and, in CTX1, the new turn remains unrelated from subcontext CTX0.

At this point, A tries to restore the intersubjectivity by planning a ``Satisfy'' action on the goal that A and B have the same interpretation context on a subpart of the dialogue. This results in the instantiation of an action
``Satisfy(A, B, inter(A, [T1], ctx) inter(B, [T1], ctx) equal(ctx, ctx))".
A plans a ``Restructure'' action and evaluates its constraints (in doing so, he applies the algorithm described in section 3.3, which, in this case, only makes him reinterpret locally T1 and relate it with T2).
A finds an alternative interpretation of his own turn T1 which explains T2 by means of an adoption relation (see CTX1 in Figure 5). So, he assigns the responsibility of the misalignment to B. In the interpretation of CTX1, which is still different from what B really has in mind, A attributes to B the belief that he wants to put a calendar on his desk. However, A has chosen the wrong hypothesis; since he is unaware of his error, he consequently formulates a request for repair (``Restructure(B, A, [T1,T2], CTX1, CTX0)''): he exploits the strategy of repeating the first turn with the addition of some more specific information about the type of calendar he wants. The new information serves to enable the partner to disambiguate between the two alternative interpretations (CTX0 and CTX0).

  
Figure 5: The reconstruction of turns T1 and T2 from Martin's point of view.

Going back to B's view of the interaction (see CTX2 in Figure 4), she successfully recognizes T3 as an unrelated turn, which is a repair following an interactional breakdown. However, in this case, it is a justified topic shift: in fact, this turn satisfies the joint goal of maintaining the intersubjectivity in the dialogue.

After B interprets T3, she finds out that the beliefs underlying her interpretation context CTX2 are inconsistent with A's request for restructuring. While B reconstructs T3 with the newly conveyed information about the desired calendar type, she understands what A thinks that she has in mind. In fact, she can recognize the ``Restructure" action in A's plan only by hypothesizing that A believes that the constraints of such an action are true: in particular, she has to reconstruct A's alternative context interpretation CTX1 (Figure 5); she can do so thanks to the fact that A has specified that the calendar should be hung on the wall, in order to contrast with the ``Put-on" interpretation (A believes that constraint ``inter(B, [T1,T2], CTX1)" is true). At the same time, B succeeds in identifying A's original intention, displayed in T1: hanging the calendar on the wall (A believes that the other constraint ``inter(A, [T1,T2], CTX1)" is true).
On the basis of all this information, B adopts the goal of resolving the complex misunderstanding: she restructures her interpretation context by changing it from CTX0 to CTX0, as a consequence of the effect of ``Restructure(B, A, [T1,T2,T3], CTX2, CTX0)" and then she notifies A that she has understood what he meant initially (turn T4).

The notification is performed by means of a ``Satisfy" action on the goal of letting A know that a ``Restructure'' action has been successfully executed.

It must be noted that B could not accept A's repair of turn T3; in fact, this would have mislead A by inducing him to believe that B's dialogue context really corresponded to CTX1. Instead, from her acknowledgment, A understands that B's ``Restructure'' action is not the one requested by him in T3: T4 ( ``Oh, you want one") means that B has realized only at this point that A wanted to keep the calendar for himself, instead of borrowing it. Now, the dialogue can go on correctly and B can provide A with the calendar to be hung on the wall.

For [Schegloff1992], Marty's controversial turn T3 is a follow-up question that specifies his intentions, rather than a repair for correcting Loes' interpretation. However, Schegloff does not exclude the other interpretation, since he admits the existence of repairs which do not have the recurrent linguistic features characterizing their canonical form: turn T3 could be a polite form of repair where, instead of correcting Loes explicitly, Marty responds to her with another question that addresses the interpretation problem. Anyway, Schegloff also points out the fact that, if Marty's turn T3 were a repair, it would be addressed to a different interpretation problem (calendar type) from the one which is addressed by Loes' fourth turn (keeping or borrowing). In our model, the double misunderstanding interpretation, that Schegloff partially admits, is not problematic: in fact, thanks to the planning context in which it is embedded, the role of T3 in triggering T4 is recognized.